Table 1.

Possible scenarios and their methodologic implications for the causal diagram representing the healthy worker survivor bias

Component*			Confounding by W(j)	W(j) affected by prior exposure	Analysis method
C1	C2	C3
1	1	1	Yes	Yes	Non-standard†
1	1	0	No	Yes	Standard‡, employment status unadjusted
1	0	1	No	Yes	Standard, employment status unadjusted
0	1	1	Yes	No	Standard, employment status adjusted
1	0	0	No	Yes	Standard, employment status unadjusted
0	1	0	No	No	Standard, employment status unadjusted
0	0	1	No	No	Standard, employment status unadjusted
0	0	0	No	No	Standard, employment status unadjusted

^*Cell entries: 0 = Absent; 1 = Present.

^†G Formula [10,51], G estimation of a structural nested model [13].

^‡Standard methods include Cox proportional hazards [27], Poisson, or logistic regression, or linear excess relative rate models [52].